杂草种子视觉不变特征提取及其种类识别

摘要

In order to study the biological stability genetic characteristics of seeds, a method of feature extraction was presented based on visual invariance.The paper analyzed the morphological characteristics and the Hu's invariant moments, and extracted 16 features of weed seeds on the basis of the analysis.These features could effectively represent the weed seeds of biological stability genetic characteristics.Experiment results showed that the extracted 16 features with visual invariant which were taken as Back Propagation (BP) neural network identification and classification of the feature sets could quickly identify the weed seeds with a high recognition rate.Therefore, the method is important in the plant quarantine and agricultural production.%为提高杂草种子的快速准确的检疫鉴定,该文基于杂草种子生物稳定遗传特征对其进行视觉不变特征提取及其种类识别的研究.首先基于形态学特征和Hu氏不变矩提取杂草种子的16个视觉不变特征,这些特征有效地表征了杂草种子的生物稳定遗传特征信息;然后基于神经网络对其进行分类.试验证明,利用提取出的具有视觉不变性的16个特征值作为BP神经网络识别分类的特征集,能快速地识别杂草种子,且具有较高的识别率.因此,该方法对植物检疫和农业生产具有重要意义.